This edition of the periodical is late… It should have been published in late August, but it is now late October. The delay reflects just how overloaded we have been. Better late than never…
One significant headache has been the Google Play requirement for apps targeting Android 15 (API 35) and above to implement edge-to-edge display, in line with its modern UI quality standards. The technical rationale includes achieving parity with iOS’s full-screen layouts and safe-area insets, and avoiding black letterboxing around app displays. A further effect has been the removal of older apps that have not been updated. In our casual games segment, there were 52,000 apps in 2024, with a further 50,000 added since then. However, this did not bring the total to 102,000: 43,000 were removed for failing to meet the required UI standards. This policy is reasonable; otherwise, the app store would rapidly become bloated. However, it also makes competition tougher for our apps.
Meeting these requirements has demanded considerable effort on our part. Our apps are native Android applications, with legacy code dating back as far as 2010. At their core, they were designed around the Google Play standards of that time, and repeated adaptations to meet successive standards have inevitably led to some less-than-optimal solutions. In total, we have submitted 34 release updates since last year, many of them covering multiple apps.
However, in terms of tough competition (alluded to above) we also analysed app placement in the Google Play store through Gemini, asking:
“what are the top 10 XXX programs on Google Play” and:
“how many XXX programs are there on Google Play that allow the user to play the game”
(testing this instead of simple Google Play store listings, which are distorted by targeted promotion campaigns)
| Game Title | Pos in list |
How many competing playable titles |
% posit all titles, where 0% is top |
Google Play Rating Free+paid |
|---|---|---|---|---|
| Chess | 2 | ~1500 to 5000 | 0.06% | 4.6 / 4.8 |
| Backgammon | 2 | ~425 to 900 | 0.30% | 4.3 / 4.4 |
| Checkers | 2 | ~425 to 900 | 0.30% | 4.4 / 4.6 |
| Euchre | 2 | ~135 to 285 | 0.95% | 4.4 / 4.4 |
| Reversi | 2 | ~200 to 400 | 0.66% | 4.5 / 4.7 |
| Hearts | 3 | ~279 to 500 | 0.77% | 4.5 / 4.7 |
| Tic Tac Toe | 3 | ~2700 to 6500 | 0.06% | 4.4 / 4.8 |
| Sudoku | 4 | ~4500 to 9500 | 0.06% | 4.5 / 4.7 |
| Spades | 5 | ~300 to 600 | 1.11% | 4.6 / 4.6 |
| Four in a Line | 7 | ~900 to 2300 | 0.44% | 4.5 / 4.7 |
| Solitaire | 8 | ~4500 to 9500 | 0.11% | 4.5 / 4.7 |
| Gomoku | 10 | ~900 to 2300 | 0.62% | 4.0 / 4.5 |
| Go | 16 | ~425 to 1350 | 1.80% | 4.4 / 5.0 |
| Gin Rummy | 17 | ~270 to 760 | 3.30% | 4.6 / 4.7 |
| Sticky Blocks | Unique | 4.5 / 4.6 | ||
| Move it! | Unique | 4.3 / 5.0 |
Despite some drop in app placement, this is still a very impressive domination of the Google Play “best casual apps” list, in the opinion of Gemini. From thousands of competing apps with the same games, AI Factory has 5 apps ranked #2 and 11 in top 8.
This edition includes an update on LLMs. We skipped this topic last year in favour of covering the World Computer Chess Championships in Spain, which we attended, so our last article, published two years ago, is already out of date. LLMs receive mixed press, but we continue to be impressed and are now heavy users. This whole periodical was prepared with the help of the LLM Claude.
One particularly impressive example concerned an article I published 11 years ago on “Interest Search” in Game AI Pro 2. It described an original search method I had developed 27 years ago but had kept under wraps for commercial reasons. I asked Claude which programs used the Interest Search idea I had published. It correctly identified the programs that used it, but also found a paper published in Japan describing a very similar principle, “realisation probability search”. This appeared long after my first use of the idea and employed a very different framework and terminology.
Neither paper referenced the other, and I suspect many academics reading both might not have made the connection. Yet Claude correctly recognised the underlying similarity. I then discussed the concept with Claude, proposing ideas that appeared in neither paper. It correctly rejected my suggestions and explained why they would not work, demonstrating an understanding that went beyond simply repeating what the papers said. It was very impressive: I felt as though I were talking to a senior academic. LLMs still leave me gobsmacked from time to time!
As ever, we attended GDC in San Francisco. Unlike in the early years, we were not spending every last minute pursuing new business through endless meetings: our small three-person team had little capacity to take on additional work. Our business was already sound, so we had the luxury of engaging with the industry as a whole, following trends and exchanging ideas with fellow practitioners. The many presentations we attended left us with notebooks full of ideas. The games industry probably changes faster than almost any other, so keeping up with the zeitgeist is essential if we are to stay on track.
The title we were developing last year is still progressing, but it has not received all the attention it needs and has consequently been delayed. The Interest Search method mentioned above plays a key role in its development. This morning, it solved a puzzle requiring 109 moves, with typically around seven legal moves available at each turn. That creates a truly gigantic number of possible continuations over that depth, beyond the practical reach of most conventional search methods.
This pending new title continues to be a very engaging game to play, play tested every day. It generates endless curious and interesting play situations, in which the player must carefully balance and trade their resources to solve the final puzzle.
October 2026
